Nevertheless, partisanship persisted: fake news warnings help briefly, but bias returns with time.

Politically oriented "fake news"-false stories or headlines created to support or attack a political position or person-is increasingly being shared and believed on social media. Many online platforms have taken steps to address this by adding a warning label to articles identified as false, but past research has shown mixed evidence for the effectiveness of such labels, and many prior studies have looked only at either short-term impacts or non-political information. This study tested three versions of fake news labels with 541 online participants in a two-wave study. A warning that came before a false headline was initially very effective in both discouraging belief in false headlines generally and eliminating a partisan congruency effect (the tendency to believe politically congenial information more readily than politically uncongenial information). In the follow-up survey two weeks later, however, we found both high levels of belief in the articles and the re-emergence of a partisan congruency effect in all warning conditions, even though participants had known just two weeks ago the items were false. The new pre-warning before the headline showed some small improvements over other types, but did not stop people from believing the article once seen again without a warning. This finding suggests that warnings do have an important immediate impact and may work well in the short term, though the durability of that protection is limited.